GSICS Annual Meeting March 06, 2019 Frascati, Italy Suomi-NPP VIIRS Reprocessing and Calibration Improvements in VNIR Bands Sirish Uprety, Changyong Cao, Wenhui Wang, Xi Shao, Bin Zhang, Slawomir Blonski, Taeyoung Choi With contributions from STAR OC Team: Junqiang Sun GSICS Annual Meeting March 06, 2019 Frascati, Italy

S-NPP VIIRS Instrument Performance VIIRS V2 Reprocessing Status Outline S-NPP VIIRS Instrument Performance VIIRS V2 Reprocessing Status Issues in operational RSB Calibration Improvements In reprocessed RSB calibration Improvements in RSBAutoCal F-factor Kalman Filter for improved gain characterization NOAA STAR Surface Roughness-induced Rayleigh Scattering (SRRS) Model DCC and SNO-x Temporal trends Summary

Background S-NPP VIIRS was launched on November 28, 2011. 14 Reflective Solar Bands (RSB, I1-I3 and M1-M11) with spectral range from 0.412 m to 2.25 m. More than 7 years of sensor data records (SDR) are available. S-NPP VIIRS has been performing well since launch Degradation is slow, RTA mirror degradation has stabilized. Overall, user feedbacks are very positive. Reprocessing aims to address remaining issues in the mission-long/life-cycle time series. Radiometric consistency in the long-term data records. Resolve major issues, such as Residual degradation in RSB SDRs after solar calibration. Bias correction

VIIRS Spectral Bands (RSB) 11 moderate bands (0.412-2.25 m). 3 imagery bands 1 Day/Night band (DNB) Calibrated using Solar Diffuser (SD) and Solar Diffuser Stability Monitor (SDSM).

Operational Calibration major Issues and Updates

Lunar Calibration Suggests Residual Degradation in Operation Calibration How to incorporate lunar calibration in VIIRS reprocessing? Lunar Solar Divergence between Lunar and Solar calibration indicates residual degradation; Lunar calibration is likely more stable. However, lunar measurements are sparse, with 9 data points per year; As a result, a combination of different validation sources including lunar is needed for near realtime correction of the residual degradation Use Kalman Filter to achieve improved calibration using SD, lunar, DCC and SNOx

Reflective Solar Band (RSB) Consistency Resolving inconsistencies due to incremental updates. For example: In the calibration equation, C0=0 update applied on May 9, 2014. I3 shows approximately 1% drop. H-Factor filter update in May 2014 1-2% increase in blue bands C0=0 update

Operational SNPP VIIRS Comparison with AQUA MODIS Before Applying SBAF After Applying SBAF Ref: https://ncc.nesdis.noaa.gov/VIIRS/VIIRS_MODIS_Intercomparison.php

S-NPP VIIRS V2 Reprocessing led calibration Improvements

S-NPP VIIRS V2 Reprocessing Improvements STAR Kalman Filter Uses Lunar/DCC/SNO; future AI capabilities Provides longterm degradation correction for VNIR Bands STAR Surface Roughness-induced Rayleigh Scattering (SRRS) Model longterm degradation correction for SWIR bands Other Improvements Improved RSBAUTOCAL F-factors re-analyzed SD/SDSM screen and BRDF LUTs (annual oscillation removed + further smoothing) Used Thuillier (2002) solar spectrum, consistent with NOAA-20 Changes radiance Updated OC F-factor based bias correction (V2) Constant bias correction M5/M7 (V1/V2)

Baseline Calibration Improvements (SD F-Factors) Baseline SDRs are generated using RSBAUTOCAL F-factors based on re-analyzed SD/SDSM screen and BRDF LUTs Annual oscillation removed + further smoothed with moving window averaging filter. Version 2 baseline SDRs show smaller annual cycles and less degradations. Bias relative to AQUA MODIS, Bias= (V-M)*100%/M Version 1 Version 2 M01 DCC Trend

Kalman Kalman Filter for Improved F-factor (Reprocessing Version 2) RSBAutoCal F Factor (Solar diffuser cal) Future plan Kalman AI based Forecasting F-Factor DCC Time Series Estimated F-Factor Time Series Lunar F Time Series SNOx Time Series Validation Prediction Features: Predict up to two weeks ahead of time for correction (used as a scaling factor); Different time resolutions (some at irregular interval) Combine input observations from different sources (subject to measurement uncertainty and noise)

Kalman Filter Based Improved F-Factors RSBAUTOCAL-F DCC-F SNOx-F Lunar-F Kalman-F

Kalman Filter Based Improved F-Factors

SNOx Trending (Before and After Kalman) Gray: Before correction Blue: After correction M1 shows dip in 2012, could be related to MODIS since this is not observed in DCC trend. Stable to 0.3% using data after 2013 only. Rest of the bands shows change less than 0.3% with 1-sigma <0.5%.

DCC (Before and After Kalman) Stability < 0.1% Stability < 0.1% Before M01 M02 Green: Before correction Black: After correction Stability: better than 0.3% for most of the bands over 5 years

SRRS Model For VIIRS SWIR Band Correction NOAA STAR Surface Roughness-induced Rayleigh Scattering (SRRS) Model Ref: Shao, X.; Cao, Changyong; Liu, T.-C. Remote Sensing 2016, 8, 254 Before DCC Trends After After using SRRS model, excellent stability in SWIR bands, <0.15% change in 5 years.

Bias Correction Factors for V2 Reprocessing RadiometricBiasCorrection (added to HDF file) 4 Elements Array added into H5 Files Radiance correction from Ocean Color group provided f factor (Calibration Method 1) Radiance correction from Kalman Filter f-factor by STAR VIIRS SDR Team (Calibration Method 2) High Gain/Low Gain RSB Single Gain: I1, I2, I3, M6, M8, M9, M10 and M11; dual gain: M1, M2, M3, M4, M5 and M7 Usage 𝑅 𝐜𝐨𝐫𝒓𝒆𝒄𝒕𝒆𝒅 =𝐶[𝑖]∙𝑅 C[i] is one of the 4 elements, R is the baseline radiance in H5 file Rcorrected is the bias corrected radiance/reflectance 1st Element Radiance Ratio between Calibration Method 1 and STAR reprocessed High Gain or Single gain values 2nd Element Radiance Ratio between Calibration Method 1 and STAR reprocessed Low Gain values (for dual gain only; or set 1 for single gain) 3rd Element Radiance Ratio between Calibration Method 2 and STAR reprocessed High Gain or Single gain values 4th Element Radiance Ratio between Calibration Method 2 and STAR reprocessed Low Gain values (for dual gain only; or set to 1 for single gain) HG F-factor ratio using OC and STAR 1 LG 2 STAR VIIRS SDR Team Kalman Filter and Constant radiometric bias for M5 and M7 3

Summary After V2 reprocessing, DCC and SNO-x trends for VNIR indicate excellent stability to within 0.3% for most of the bands RSB baseline SDRs calibrated using improved (removed annual oscillation and smoothed) SD calibration. Radiometric bias correction is provided to users for each granule Kalman model derived coefficients (using lunar, DCC, SNO and SD) integrated with M5 and M7 absolute bias for VNIR bands. Option to use OC derived gain trends Surface Roughness-induced Rayleigh Scattering (SRRS) Model for SWIR bands As of 2/28/2019, completed VIIRS SDR V2 reprocessing for 2016. Completion timeline for VIIRS SDR V2 reprocessing for 2012/01-2017/03 is by the end of July (on schedule) Version 2 provides the solid basis for further future reprocessing. In most cases, future reprocessing will be simply based on further improved bias correction without fundamentally reprocessing RDR to SDR. V2 reprocessing provides more accurate, stable, and consistent radiometric calibration, meeting all user needs

Backup

VIIRS RSB/TEB/DNB Band Center wavelength [um] Spatial reso. at nadir [m] Gain M1 0.412 750 High/Low M12 3.70 Single M2 0.445 M13 4.05 M3 0.488 M14 8.55 M4 0.555 M15 10.76 M5 0.672 M16 12.01 M6 0.746 I1 0.640 375 M7 0.865 I2 M8 1.24 I3 1.61 M9 1.38 I4 3.74 M10 I5 11.45 M11 2.25 DNB 0.7 H/M/L† *The shaded area indicates thermal emissive bands. †High, mid, and low gain states

VIIRS SWIR Band Corrected with SRRS Model Percent Correction Factor Percent Correction Factor DCC Mode DCC Mode After Correction Slope < 0.15% over 5 years Slope < 0.1% over 5 years After Correction